- Glasgow-haskell-users

Confusing Typing Problem
by Bernd Brassel 18 Nov '05

18 Nov '05

I am a bit confused by GHC behaviour for the following variants of a program: data A = A (Int -> A) f = A g g _ = f h = g 'c' ghc: > Error: > Couldn't match `Int' against `Char' The problem is that a small change ommits the error: data A = A (Int -> A) f = A g where g _ = f h = g 'c' ghc: > Ok, modules loaded But I really need the functions to be top-level, thus I added type signatures: data A = A (Int -> A) f :: A f = A g g :: a -> A g _ = f h :: A h = g 'c' ghc: > Ok, modules loaded Fine, but what I really need, is this: data A = A (Int -> A) f :: A f = A g g :: Ord a => a -> A g _ = f h :: A h = g 'c' > Contexts differ in length > The signature contexts in a mutually recursive group should all be > identical > ... > Couldn't match `Int' against `Char' Again the problem does not occur for local functions: data A = A (Int -> A) f :: A f = A g where g :: Ord a => a -> A g _ = f h :: A h = g 'c' Why is ghc treating local and global functions differently? Is there really a difference in decidability of the Polymorphic Type Inference? And regarding the "contexts differ in length", I read a comment in "Typing Haskell in Haskell": > a throw-away comment specifying that all explicit type signatures > in a binding group must have the same context up to renaming of variables > [10,Section 4.5.2]. This is a syntactic restriction that can easily > be checked prior to type checking. Our comments here, however, suggest > that it is unnecessarily restrictive. Why does ghc still use that unnecessary restriction? Cheers Bernd

1 0

RE: Compiling GHC in Windows, Mingw
by Simon Marlow 17 Nov '05

17 Nov '05

On 17 November 2005 06:15, Esa Ilari Vuokko wrote: > I tried to compile ghc from cvs head in windows, hosting and > targetting mingw. When I got ghc to compile, it crashes while > parsing (I putStrLn-debugged) package.conf, but only from installed > position. > > Basically, somewhere inside ParsePkgConf.loadPackageConfig GHC crashes > with access violation. With very few test runs, it appears that > simplifying the file helps, but only case I got it not to crash was > one package in file, rts, and empty ldOptions. > > Maybe I went wrong somewhere, any pointers appreciated! > > What I did: > Current msys, msysdtk, binutils, win32api, gcc-core. > cvs checkout fpconfig and in there ghc, libraries, alex and happy > autoreconf > ./configure --host=i386-unknown-mingw32 > --with-gcc=d:/tools/msys/mingw/bin/gcc.exe > make > make install I haven't seen this, but I'm just updating my Windows build to try to reproduce. Does it happen with the stage1 or stage2 compiler? What version of GHC are you bootstrapping with? Cheers, Simon

2 2

RE: kind inference
by Simon Peyton-Jones 17 Nov '05

17 Nov '05

Yes it's deliberate. (Turned out to be easy and convenient.) Yes it goes beyond H98, so GHC types a few too many programs even in H98 mode. I guess I should document it. Simon | -----Original Message----- | From: glasgow-haskell-users-bounces(a)haskell.org [mailto:glasgow-haskell-users- | bounces(a)haskell.org] On Behalf Of Ross Paterson | Sent: 17 November 2005 11:31 | To: glasgow-haskell-users(a)haskell.org | Subject: kind inference | | The Haskell 98 Report says kind inference is to be done over dependency | groups with polymorphic kinds defaulted to *, and hence the following | example (from 4.6) is illegal: | | data Tree a = Leaf | Fork (Tree a) (Tree a) | | type TreeList = Tree [] | | GHC 6.4 seems to perform kind inference across all the data, newtype, | type and class declarations of a module before defaulting, and so accepts | the above. An improvement, I think, but unadvertised. Was it intentional? | | _______________________________________________ | Glasgow-haskell-users mailing list | Glasgow-haskell-users(a)haskell.org | http://www.haskell.org/mailman/listinfo/glasgow-haskell-users

2 1

kind inference
by Ross Paterson 17 Nov '05

17 Nov '05

The Haskell 98 Report says kind inference is to be done over dependency groups with polymorphic kinds defaulted to *, and hence the following example (from 4.6) is illegal: data Tree a = Leaf | Fork (Tree a) (Tree a) type TreeList = Tree [] GHC 6.4 seems to perform kind inference across all the data, newtype, type and class declarations of a module before defaulting, and so accepts the above. An improvement, I think, but unadvertised. Was it intentional?

1 0

Compiling GHC in Windows, Mingw
by Esa Ilari Vuokko 17 Nov '05

17 Nov '05

Hi! I tried to compile ghc from cvs head in windows, hosting and targetting mingw. When I got ghc to compile, it crashes while parsing (I putStrLn-debugged) package.conf, but only from installed position. Basically, somewhere inside ParsePkgConf.loadPackageConfig GHC crashes with access violation. With very few test runs, it appears that simplifying the file helps, but only case I got it not to crash was one package in file, rts, and empty ldOptions. Maybe I went wrong somewhere, any pointers appreciated! What I did: Current msys, msysdtk, binutils, win32api, gcc-core. cvs checkout fpconfig and in there ghc, libraries, alex and happy autoreconf ./configure --host=i386-unknown-mingw32 --with-gcc=d:/tools/msys/mingw/bin/gcc.exe make make install (I had to change some things, as described on mail to bugs-list, so few repeated make's) I will try clean rebuild later, but I feel doubtful. Best regards, --Esa

1 0

RE: [Haskell-cafe] GHC ignores RTS options
by Simon Marlow 16 Nov '05

16 Nov '05

On 16 November 2005 13:24, Tomasz Zielonka wrote: > On Wed, Nov 16, 2005 at 01:09:07PM -0000, Simon Marlow wrote: >>> BTW - it would be nice if it was easier to override the default RTS >>> options when compiling the program. Perhaps a nice task for a newbie >>> GHC developer? >> >> Already possible, see: >> >> http://www.haskell.org/ghc/docs/latest/html/users_guide/runtime-control. >> html#rts-hooks > > I was aware of this, and I should write about it. > >> Admittedly it's a bit more fiddly than a command-line flag to GHC, >> though. > > And because of this I wrote that it would be nice if it was easier :-) Sorry, I misread the message. > I've bitten several times by the default values of RTS options, > especially the stack size limit (-K). But it was always easier to > set GHCRTS or use +RTS -RTS, so I never used the rts-hooks. > > Do you think I could do this? I am not sure how difficult it would be. Sure, it shouldn't be too hard. You need to generate a .c file containing the ghc_rts_opts setting, and then compile & link it into the final binary. It's just a matter of plumbing the various bits together in GHC's front end. (it might be nicer to generate Cmm and then push this through the back end, to avoid the dependence on the C compiler, though). Cheers, Simon

2 1

RE: GHC 6.4.1 on Solaris x86
by Simon Marlow 16 Nov '05

16 Nov '05

On 16 November 2005 13:18, Duncan Coutts wrote: > On Wed, 2005-11-16 at 12:59 +0000, Simon Marlow wrote: >> On 14 November 2005 04:17, Ben Horsfall wrote: >> >>> Is there a binary package for GHC 6.4.1 on Solaris x86 (Sun OS 5.9)? >>> >>> Building from source fails thus: >>> >>> ../../ghc/compiler/ghc-inplace -optc-O -optc-Wall -optc-W >>> -optc-Wstrict-prototypes -optc-Wmissing-prototypes >>> -optc-Wmissing-declarations -optc-Winline -optc-Waggregate-return >>> -optc-Wbad-function-cast -optc-I../includes -optc-I. -optc-Iparallel >>> -optc-DCOMPILING_RTS -optc-fomit-frame-pointer -H16m -O -O2 -static >>> -I. -#include Prelude.h -#include Rts.h -#include RtsFlags.h >>> -#include RtsUtils.h -#include StgRun.h -#include Schedule.h >>> -#include Printer.h -#include Sanity.h -#include STM.h -#include >>> Storage.h -#include SchedAPI.h -#include Timer.h -#include Itimer.h >>> -#include ProfHeap.h >>> -#include LdvProfile.h -#include Profiling.h -#include Apply.h >>> -fvia-C >>> -dcmm-lint -c StgCRun.c -o StgCRun.o >>> Assembler: StgCRun.c >>> "/tmp/ghc14327.s", line 18 : Illegal mnemonic >>> "/tmp/ghc14327.s", line 18 : Syntax error >>> "/tmp/ghc14327.s", line 18 : Illegal mnemonic >>> make[2]: *** [StgCRun.o] Error 1 >> >> There isn't a binary for 6.4.1 that I know of. You should be able to >> follow the porting instructions in the Building Guide to get an >> unregisterised version going (get the version from CVS, there were >> some recent modifications). > > My department uses 6.4.1 on Solaris 9 and it works nicely. They built > it from source using version 6.2.2 (for which there are binaries I > believe). I don't think it's necessary to do a fresh port. Ah good. Any chance of building a binary dist I can put up? Cheers, Simon

2 1

RE: [Haskell-cafe] GHC ignores RTS options
by Simon Peyton-Jones 16 Nov '05

16 Nov '05

I'd love to see this done. It's been requested several times. Simon | -----Original Message----- | From: glasgow-haskell-users-bounces(a)haskell.org [mailto:glasgow-haskell-users- | bounces(a)haskell.org] On Behalf Of Simon Marlow | Sent: 16 November 2005 15:48 | To: Tomasz Zielonka | Cc: Dusan Kolar; GHC-users; Haskell-Cafe(a)haskell.org | Subject: RE: [Haskell-cafe] GHC ignores RTS options | | On 16 November 2005 13:24, Tomasz Zielonka wrote: | | > On Wed, Nov 16, 2005 at 01:09:07PM -0000, Simon Marlow wrote: | >>> BTW - it would be nice if it was easier to override the default RTS | >>> options when compiling the program. Perhaps a nice task for a newbie | >>> GHC developer? | >> | >> Already possible, see: | >> | >> | http://www.haskell.org/ghc/docs/latest/html/users_guide/runtime-control. | >> html#rts-hooks | > | > I was aware of this, and I should write about it. | > | >> Admittedly it's a bit more fiddly than a command-line flag to GHC, | >> though. | > | > And because of this I wrote that it would be nice if it was easier :-) | | Sorry, I misread the message. | | > I've bitten several times by the default values of RTS options, | > especially the stack size limit (-K). But it was always easier to | > set GHCRTS or use +RTS -RTS, so I never used the rts-hooks. | > | > Do you think I could do this? I am not sure how difficult it would be. | | Sure, it shouldn't be too hard. You need to generate a .c file | containing the ghc_rts_opts setting, and then compile & link it into the | final binary. It's just a matter of plumbing the various bits together | in GHC's front end. | | (it might be nicer to generate Cmm and then push this through the back | end, to avoid the dependence on the C compiler, though). | | Cheers, | Simon | _______________________________________________ | Glasgow-haskell-users mailing list | Glasgow-haskell-users(a)haskell.org | http://www.haskell.org/mailman/listinfo/glasgow-haskell-users

1 0

RE: [Haskell-cafe] GHC ignores RTS options
by Simon Marlow 16 Nov '05

16 Nov '05

On 16 November 2005 08:12, Tomasz Zielonka wrote: > On Wed, Nov 16, 2005 at 08:53:18AM +0100, Dusan Kolar wrote: >> I wonder whether anybody could advise me how to pass >> correctly RTS options to ghc? I've tried > > You have to pass these options to the program, not > to GHC. > > BTW - it would be nice if it was easier to override the default RTS > options when compiling the program. Perhaps a nice task for a newbie > GHC developer? Already possible, see: http://www.haskell.org/ghc/docs/latest/html/users_guide/runtime-control. html#rts-hooks Admittedly it's a bit more fiddly than a command-line flag to GHC, though. Cheers, Simon

2 1

RE: GHC 6.4.1 on Solaris x86
by Simon Marlow 16 Nov '05

16 Nov '05

On 14 November 2005 04:17, Ben Horsfall wrote: > Is there a binary package for GHC 6.4.1 on Solaris x86 (Sun OS 5.9)? > > Building from source fails thus: > > ../../ghc/compiler/ghc-inplace -optc-O -optc-Wall -optc-W > -optc-Wstrict-prototypes -optc-Wmissing-prototypes > -optc-Wmissing-declarations -optc-Winline -optc-Waggregate-return > -optc-Wbad-function-cast -optc-I../includes -optc-I. -optc-Iparallel > -optc-DCOMPILING_RTS -optc-fomit-frame-pointer -H16m -O -O2 -static > -I. -#include Prelude.h -#include Rts.h -#include RtsFlags.h -#include > RtsUtils.h -#include StgRun.h -#include Schedule.h -#include Printer.h > -#include Sanity.h -#include STM.h -#include Storage.h -#include > SchedAPI.h -#include Timer.h -#include Itimer.h -#include ProfHeap.h > -#include LdvProfile.h -#include Profiling.h -#include Apply.h -fvia-C > -dcmm-lint -c StgCRun.c -o StgCRun.o > Assembler: StgCRun.c > "/tmp/ghc14327.s", line 18 : Illegal mnemonic > "/tmp/ghc14327.s", line 18 : Syntax error > "/tmp/ghc14327.s", line 18 : Illegal mnemonic > make[2]: *** [StgCRun.o] Error 1 There isn't a binary for 6.4.1 that I know of. You should be able to follow the porting instructions in the Building Guide to get an unregisterised version going (get the version from CVS, there were some recent modifications). Cheers, Simon

2 1